Introduction

Objective:

• Assess the relationships between habitat quality and biodiversity in a highly fragmented and human-dominated forested landscape in Massachusetts, through a geospatial analysis lens, integrating spatial analysis, remote sensing, and soundscape ecology.

• Understand the relationships between community acoustic and habitat configuration.

Package: NatureWave

• Package which analyzes processed Sound Index data for NDSI (Normalized Difference Sound Index)

• Assesses the correlations between sound indices and ecological variables of interest taking into account different landscape metrics for Massachusets at different scales of influence.

• Evaluates the importance of ecological variables at multiple scales and predicts NDSI values accross the entire study area.

Acoustic Indices

Acoustic indices allow an aggregation of acoustic signals to represent the overall acoustic community diversity of the landscape.

The Normalized Difference Soundscape Index (NDSI) characterizes soundscape properties as such to allow the identification of human influence of the landscape through the partitioning of soundscapes into biophony (sounds from biodiversity) and anthrophony (sounds from humans).

This partitioning allows the classification of a particular location based on its ecological and anthropic characteristics.

NDSI = (biophony - anthrophony)/ (biophony + anthrophony)

This project provides a less costly-reproducible approach to measurement and analysis and contributes to the applications of remote sensing technologies in conservation planning and monitoring of habitat quality.

Data

• Conservation Assessment and Prioritization System (CAPS) Landscape Metrics

Statewide Massachusetts Assessment: November 2011 http://jamba.provost.ads.umass.edu/web/caps2011/CAPS2011data.htm

• Dr. Florencia Sangermano collected recordings at 11 different sites in Central Mass. over the course of 3 months between June and August.

## Reading layer `sample_sites_projected' from data source `C:\Users\jyarlequeipanaque\Documents\R\win-library\3.6\naturewave\extdata\sample_sites_projected.shp' using driver `ESRI Shapefile'
## Simple feature collection with 11 features and 5 fields
## geometry type:  POINT
## dimension:      XY
## bbox:           xmin: 146419.9 ymin: 869709.6 xmax: 178745.5 ymax: 928660.8
## proj4string:    +proj=lcc +lat_1=41.71666666666667 +lat_2=42.68333333333333 +lat_0=41 +lon_0=-71.5 +x_0=200000 +y_0=750000 +datum=NAD83 +units=m +no_defs

## Reading layer `AOI' from data source `C:\Users\jyarlequeipanaque\Documents\R\win-library\3.6\naturewave\extdata\AOI.shp' using driver `ESRI Shapefile'
## Simple feature collection with 1 feature and 0 fields
## geometry type:  POLYGON
## dimension:      XY
## bbox:           xmin: -72.30321 ymin: 42.02124 xmax: -71.53417 ymax: 42.69707
## CRS:            4326

Methodology

Focal Mean

# # Focal Mean using circular focalWeight at a buffer of 2500 meters
# 
# #2500 Buffers
# 
# fw_sum_2500 <- lapply(landcover_list,function(x){
#   fw <- focalWeight(x, 2500, type='circle')
#   })
# 
# 
# #Runs focal at a buffer weight of 2500 m radius
# 
# radius3 <- 2500
# lc2500 <- lapply(1:4, function(x) {  # x <- 1
#   f <- paste0("../external/data/Amherst_CAPS2011", names(lc_stack)[x], "_", radius3, ".tif")
#   r <- focal(lc_stack[[x]], w = as.matrix(fw_sum_2500[[x]]), filename = f, overwrite = TRUE)
#   return(r)
# })
# lc2500_stacked <- stack(lc2500)  # Stack
# 
# 
# # lc2500_brick <- brick(lc2500_stacked) # Brick
# 
# # Plotting out the results
# tit3 <- "2500 (m) Buffer"
# par(mfrow = c(2, 2), mar = c(.5, 0, 2, 6))
# lapply(1:4, function(x){
#   plot(lc2500_stacked[[x]], axes = FALSE, box = FALSE, main = toupper(c(tit3, gsub(".tif", "", basename(f[x])))))
#   plot(st_geometry(sample_points_f), add = TRUE)
# })

Random Forest

# Create RandomForest Model
set.seed(100)
rf_2500 <- randomForest(NDSI ~., data=DS_2500_scaled, mtry= 2, importance = TRUE, na.action=na.omit)

randomForest::varImpPlot(rf_2500, main = "Variable Importance at 2500 buffer level")

Predictive modeling

# Use raster::predict to predict NDSI values across the landscape
ndsi_pred <- raster::predict(object = new_stack, model=rf_2500, type = 'response', index = 5)
ndsi_pred

Results from Preliminary work

Linear Modeling

Results from Focal Mean

## [[1]]
## NULL
## 
## [[2]]
## NULL
## 
## [[3]]
## NULL
## 
## [[4]]
## NULL

Results from Random Forest

Variable Importance plot

## [1] 8 5

## [1] 3 5

## 
## Call:
##  randomForest(formula = NDSI ~ ., data = DS_2500_scaled, mtry = 2,      importance = TRUE, na.action = na.omit) 
##                Type of random forest: regression
##                      Number of trees: 500
## No. of variables tried at each split: 2
## 
##           Mean of squared residuals: 0.02077052
##                     % Var explained: 57.08

##               %IncMSE IncNodePurity
## Imperv           6.69          0.10
## Connectedness    5.75          0.13
## Structure        5.54          0.11
## Traffic          9.61          0.11

Results from ¨Predictive Modeling

NDSI Predidictive surface

Conclusion and Discussion

• We assessed the relationships between habitat quality and biodiversity in a highly fragmented and human-dominated forested landscape in Massachusetts.

• We examined four ecological variables: imperviousness, connectedness, vegetation structure and traffic to understand their relationship to sound indices.

• We used mutiple buffers to examine this relationship. The scale of best influence for NDSI values is at 2500 meters buffer from the sampling location.

• Imperviousness and Traffic are the most important ecological variables at predicting NDSI values across the study area.

Future work:

• Test further variables such as ecological integrity and micro climate alterations.

• Validate the predictive model based on an independent acoustic sample collected during May to July in the near future, following the same coection stantards as before.

• Assess the capacity of models to extrapolate relationships between habitat structure and acoustic diversity.